Skip Navigation CECS Home | ANU Home | Search ANU | HORUS | Staff Home

PhD research opportunities at RSISE

Department of Information Engineering

Our expertise and research interests are in handling uncertain and complex systems. The department undertakes basic research as well as applied research in robust autonomous robots, car automation and mobile communications. Key scientific challenges in five areas of research in the department include robotics, control systems, computer vision, systems optimization and wireless communications.

• Robotic Systems Laboratory

Humanoid Walking Algorithms

How can we fix deficiencies in present generation humanoids? Deficiencies like clumsiness, a lack of versatility, and a small repertoire of movement restrict the uses that machines can be put to, and they can not operate as smoothly as we would like them to. A PhD project in humanoid walking algorithms aims to fix these deficiencies.

Contact: Dr Roy Featherstone

Robot Kinematics, Dynamics and Control

These projects include theory, algorithms, software, and the design and construction of experiments and experimental robotic devices for improving the physical motion skills of robots.

Contact: Dr Roy Featherstone

Small Submersibles

The KAMBARA project is one of the few full scale underwater autonomous vehicles in the southern hemisphere. It is the result of research into coordinating control mechanisms and communication schemes like those exhibited by schooling fish. Research ultimately aims to create schools of autonomous underwater vehicles that can behave in the same way as schooling fish, so that 'pods' of underwater vehicles can move and localise without external control.

Contact: Dr Uwe Zimmer

• Systems and Control

Cooperative Control and Formations

How do swarms of agents, like fish or unmanned autonomous vehicles, manage to move in a formation, split a formation, merge formations, follow a leader, change formation and so on without colliding, and without a master device to drive them? Problems like this are the fascinating province of cooperative control. A PhD project aims to develop procedures for the design of system architectures to allow formation maintenance, change, merge and splitting of groups of autonomous vehicles.

Contact: Professor Brian Anderson

Reduced Complexity Robust Control

Systems that vibrate violently (i.e. resonate) when given small excitations are known as lightly damped systems. These systems include flexible mechanical structures, resonant electrical or electronic circuits, and so forth. These kind of systems are notoriously hard to control because very small imprecisions in modelling these resonant behaviours induces very large variations of the system model. Recent advances in robust control theory will enable us to tackle this problem systematically. There are numerous real-world application areas which would directly benefit from such a study - e.g. control of large space structures, flexible robotic arms, certain telecommunication networks and electronic circuits, etc.

Contact: Dr Alexander Lanzon

Iterative Identification and Controller Redesigns

Imagine that the real physical plant which we need to control is unknown to us, but we have a very crude plant model (e.g. we only know its low frequency gain). Suppose furthermore that we have designed a very basic controller (e.g. a simple constant gain controller) on the basis of the crude plant model that stabilises both the real unknown physical plant and its crude model. While this initial basic controller is operating on the real unknown physical plant, we could run an identification experiment to obtain a better plant model. Then we could use this better plant model to design a controller that is also better for the real unknown physical plant. Plugging in this newly designed controller onto the real unknown plant, we could repeat this iterative procedure endlessly to obtain excellent plant models and an optimally performing controller. However, there are some serious theoretical pitfalls that require proper analysis in order to ensure that this procedure always results in monotonic improvement of performance on the real unknown physical plant.

Contact: Dr Alexander Lanzon

Performance and Robustness Tradeoffs

Any practicing control engineer will tell you that the most difficult part of a control design process is to optimally tradeoff performance and robustness requirements. This is often done via a long and tedious trial and error process based primarily on engineering judgment and intuition. In this project, we will optimally tradeoff performance and robustness requirements using optimisation theory. This will provide designers with an important and efficient tool that enables them to concentrate on more fundamental design issues.

Contact: Dr Alexander Lanzon

Robust Control Theory

Mathematical models that perfectly describe physical systems are a rare commodity that is typically unavailable in application. Furthermore, noises and unexpected disturbances corrupt measured signals in practice. Robust control theory deals with these discrepancies in a coherent feedback framework. There are several technical research problems that necessitate investigation in this area.

Contact: Dr Alexander Lanzon

Complex Networks:   Dynamics and Structure

Numerous fundamental questions must be addressed about the connections between network structure and (nonlinear) dynamic properties including stability, bifurcations, controllability and observability.

Contact: Professor David Hill

Optimal Network Planning

The project will develop a dynamic optimization framework for planning networks according to multi-objective criteria including dynamic specifications across multiple time-scales.   Applications to energy and information networks will be studied.

Contact: Professor David Hill

Global and Coordinated Control

The term 'global control' refers to the intention to control wide-area and nonlinear phenomena in various operating conditions. Global control is enhanced by optimization to achieve co-ordination of control capability scattered across the system in order to address normal and emergency conditions.

Contact: Professor David Hill

Switched Systems and Networks

A theory of switched systems is currently being developed. This project aims to make this more specific to network systems and provide a theoretical basis for dynamic analysis and control.

Contact: Professor David Hill

Pattern-based Control

The example of high-performance sports suggests that control should be based on iterative recognition of trajectory patterns with responses consisting of a combination of open-loop and closed-loop actions. With ideas from learning control in a hybrid state-space, it appears possible to implement effective control of complex situations.

Contact: Professor David Hill

Adaptive Networks

A theory of 'adaptive networks' will be developed, ie network systems where the parameters - particularly those related to structure and link capacities - can be switched and/or tuned.

Contact: Professor David Hill

Power Networks and Control

The project will emphasise two directions following on from the more basic work in other projects:

• how to expand and operate power networks optimally with respect to economy and security;
• how to implement global control architectures which allow effective coordination.

Contact: Professor David Hill

• Computer Vision (link)

Synthesis of Stereoscopic Movie from Conventional Monocular Video Clips

In order to provide material for 3-dimensional television displays, methods are required for producing 3-dimensional video material from existing 2-dimensional video, such as old films. This project seeks to develop automatic and interactive methods for 3-dimensionalizing video for this purpose. What is required is to create a synthetic stereo pair from a single video sequence. This involves synthesis of a matching image, which in conjunction with the original image will give an illusion of 3-dimensions. Methods of geometric computer vision and “structure from motion” will be used to do this. By analysis of the motion of objects in the video sequence, and determination of the motion of the camera, the structure of the scene can be determined. Once the geometry of the scene is understood, a stereo pair of images can be produced.

Contact: Professor Richard Hartley

Medical Image Analysis

Computer vision and image understanding are an important part of modern medicine. Software systems are used to help clinicians to diagnose diseases, screen for abnormal conditions, and visualize body anatomy. Our particular interest is in using computer vision techniques to help in two areas: ophthalmology and cancer detection. Diseases of the eye, such as diabetic retinopathy, or glaucoma can lead to blindness if untreated. Screening of at-risk patients can detect these diseases at an early stage. For instance, in glaucoma, pressure in the eyeball causes damage to the retina and optic nerve. Stereoscopic imagery can be used to determine the degree of deformation of the eyeball due to pressure, and hence determine the severity of the condition. The required computer tools include stereoscopic analysis of image pairs, detection of abnormal features in the retina, retinal image alignment and matching.

Screening for colon cancer is invasive costly and uncomfortable. A developing technique, virtual colonoscopy seeks to replace this method with a method based on analysis of Computer Tomographic (CT) images. Challenges involve the detection and visualization of the colon wall as an aid to interactive screening.

Contact: Professor Richard Hartley

Application of the Hyperspectral Camera to Vision Research for Near-Range Applications

A hyperspectral camera produces images in which at each pixel (image position) a complete visible range spectrum is captured. Instead of the usual red-green-blue bands captured by a normal digital camera, as many as 256 image bands from the visible range are captured. This extra information makes it possible to derive much More info...rmation from the image about the material properties of each object type in the scene. As an example, it is possible to distinguish chlorophyll-A from chlorophyll-B in plants, just on the spectrum of reflected light. This project seeks to develop applications of this technology in analyzing images taken with a hyperspectral camera. Important problems include segmentation of the image into regions of similar spectral characteristics, determination of efficient storage methods for hyperspectral images, material determination and detection of objects based on their spectra, geometric correction of images suffering from motion distortion, and alignment of multispectral images with images of other modalities.

Contact: Professor Richard Hartley

Application of Mathematical Theories to Geometric Computer Vision Problems

The mathematical analysis of image sequences involves the use of Projective Geometry techniques. The usual camera model is that of a pinhole camera, which may be simply modeled in terms of a linear projection of projective 3-space to projective 2-space. In the last decade, increasingly sophisticated use of projective geometry has led to methods of scene reconstruction from a moving camera observing a stationary scene. Algorithms for automatic calibration of the camera and reconstruction of the scene have been developed. Analysis of the failure and critical configurations is partially complete. Avenues for further work include the investigation of moving scenes observed with moving cameras (dynamic scenes), and exploitation of specific configurations in the scene, such as scene planes, curved edges and curved objects.

To undertake research in this area, a certain mathematical sophistication is necessary. A firm basis of linear algebra and the ability to become familiar with Projective Geometry are essential. Familiarity with some numerical methods would also be a plus. This project is suitable for a student with a taste for mathematics.

Contact: Professor Richard Hartley

Video Synopsis

A system is proposed that will take video input from a video camera and make a summary video. The envisaged scenario is one in which surveillance cameras are used to process large amounts of video, most of which is of no or little interest. To store all the video provided by the sensor would require vast amounts of storage, and would also be far more than a human operator would want to examine. The envisaged system would retain and store only video footage that contains interesting material.

We imagine an unsupervised video sensor (camera) placed in some location, gathering video information continuously. Such a location may be in a public place where not much activity occurs, in a house or in some remote location. Thus, we envisage placing such sensors both at indoors and outdoor locations. Typically very little activity will occur, and there will be no need to retain, or transmit most of the frames. The system will need to distinguish between normal and uninteresting activity, such as waving of trees or motion of clouds. Such decisions can be made on the basis of learning the normal variance of the scene. At a more sophisticated level, particularly related to indoor surveillance, the system can learn to distinguish human subjects, and their normal behaviour, only flagging unusual actions, such as people climbing through windows, or lying on the floor.

Contact: Professor Richard Hartley

Interactive model-based scene understanding and modelling frommultiple images or video

Many of the geometric problems of determining the structure of a set of points from multiple images have been solved. However, the problem of correct modelling of more complex scenes to produce a complete 3-dimensional model remains. To accomplish this task, we need to recognize specific geometric primitives and generic objects in the scene. Thus, recognition of planes, such as the ground plane or the sky allow it to be correctly separated from the rest of the scene and modeled accordingly. Correct detection and modelling of curved surfaces will help to create shape-specific models. Finally, recognition of objects such as trees, buildings and other common outdoor objects will allow for more faithful graphic models to be produced. The outcome would be a capability to generate better graphical models of natural scenes, leading to accurate generation of novel views.

Contact: Professor Richard Hartley

Computer Vision and Machine Learning

We seek to apply methods of machine learning, particularly kernel-based learning techniques to the solution of problems in computer vision. Specific problems addressed are object recognition and localization (position determination). This project involves collaboration with a European project named LAVA, which aims at applying such methods to applications in mobile computing. Thus, images taken with a hand-held camera/digital assistant device can be used as an input device for providing immediate assistance. For example, recognition of the location of the image can be used to bring up local information about the environment. For instance, an image taken inside a shopping mall can bring up a map of the mall, with directions to find a desired shop, from the recognized location where the present image was taken.

Contact: Professor Richard Hartley

Constraints-based scenes understanding and modelling from video tapes

Our challenge is to develop a novel method for building realistic 3D graphic models of very complex scenes from multiple images or video tapes. Most existing methods mainly follow Marr‚s philosophy, which believes that 3D information recovery is the first step of vision perception. They seek methods of constructing fully-automated vision machines. However the results are not satisfactory. Instead, we seek to build a machine vision system that can augment our eyes, rather than replacing our eyes. Therefore, human knowledge will drastically reduce the complexity and increase the performances of a 3D vision modelling system. Here, we mainly want to exploit the observed constraints contained in visual scenes, such as coplanarity, orthogonality and other geometric relations which are present in man-made structures. We believe that interaction between humans and computers will help to realize a more practical and more efficient vision machine. 

Contact: Dr Hongdong Li

Probabilistic-visual-learning-based 3D object recognition,localization and tracking

Learning is an essential ability of any intelligent system, needless to say also of our humans. Machine Learning methods should play a more important role in intelligent vision systems. However, how to build such a system that can learn from examples and errors and thus grow its intelligent capability is still an open problem, and also a big challenge as well as a big opportunity for computer vision research. This project aims at developing such a visual learning method, which is applicable for various vision tasks such as image segmentation, 3D object recognition, localization and tracking, as well as 3D reconstruction, etc. Our idea is to augment the method of probabilistic visual learning for object representation (Pentland) by active appearance model techniques. Not only 3D geometries, but also pose/position and kinemic/dynamic properties of moving objects are described in a probabilistic framework. As a result, humans' a priori knowledge can be incorporated in a natural way and in an early stage of learning, and thus increase the robustness of vision systems.

Contact: Dr Hondong Li

Pattern recognition and Vision Augmentation in a pen-based (tablet PC) computing environment

The Tablet PC is a new kind of computer, representing an evolutionary step in the development of the laptop computer used today in mobile computing. It delivers new and easy ways to interact between humans and the computer, and vastly extending the ways in which people will work and enjoy their PCs. On a Tablet PC, users can write/draw directly on its wide screen and save electronic notes in their own handwriting/painting or they can be transformed and utilized in more compact forms through a highly accurate recognition engine. Nowadays there are many character recognition engines available for tablet PCs. The real challenge is to extend such engines to non-text applications. In this research task, we are going to develop several powerful and heterogeneous recognition algorithms that are able to recognize hand-drawn diagrams/sketches/graphs/commercial charts, and mathematical expressions. We can even make a panoramic view or model 3D scenes simply by doodling on the screen. We see the highest potential for success in using a technique called semantic-syntactic structural pattern recognition method which has been proven to be a very powerful recognition method.

Contact: Dr Hongdong Li

• Systems Optimization

Solving Optimization Problems

Many problems in engineering involve the solution of optimization problems. This translates into finding the best solution or best design subject to various constraints. The challenge is to make a fundamental impact on how various important classes of optimization problems are solved in practice. While this work is often motivated by problems from areas like systems and control, signal processing, and telecommunications, it is also often of interest to researchers and practitioners in other fields.

Contact: Dr Robert Orsi

Optimizing Vision and Robotic Systems

Achieving better performance of existing engineering systems, or making things happen that were not previously possible in real time, is a challenging field of research. The systems can be vision systems, robotics systems or communications systems. The aim is to achieve techniques from areas of optimization, signal processing and control which are elegant and definitive, perhaps opening up the possibility to solve many specific problems.

Contact: Professor John Moore

Applying Mathematical Systems Theory to Real World Engineering Problems

These problems include optimisation of communication systems, designing codes for communication, perspective vision systems, reliability analysis of complex systems, or general representation and filtering problems. Depending upon student background and interest, the aim is to come up with new and better solutions to existing problems. For example, one solution could be using algorithms that allow researchers to solve wider classes of problems than those that currently exist.

Contact: Dr Jochen Trumpf

• Wireless Communications

Broadband Wireless Challenge

Wireless communication is about connecting people because whether communication is from person to person, person to machine, or machine to person, someone will be involved in the communication link. The major challenge for telecommunications is how to communicate without wires because no-one is about to carry around metres of cabling in order to communicate better, faster, cheaper and more safely with someone (or something) else. So, how do you get the plethora of high speed data circulating in the Internet to a person with a mobile terminal? We don't know. Research hopes to find the answer. Enabling lifestyle by communicating without wires is a fascinating part of fundamental research into wireless communications.

Contact: Professor Rodney Kennedy

CDMA User Tracking and Quality of Service

Mobile phone users in the Code-Division-Multiple-Access environment demand high bandwidth and low access cost. This encourages network operators to provide better or additional services without recourse to additional bandwidth, and motivates tracking of a single user given only their CDMA signals, across multiple cells. It is an open problem to show what fundamental limits apply for channel prediction and tracking of a single user, and to extend this to multiple users. There is a large benefit for users and operators in developing algorithms which approach these limits. The development of online auctions for quality of service (QoS) Guarantees across multiple cells is an immediate application of an accurate tracking system. In our research project we ask questions like: If a user wishes to download a large file to a mobile handset, should she pay extra to download immediately, or wait until she moves to a nearby cell? How should premium bandwidth be allocated to new entrants to a given cellular region?

Contact: Dr Leif Hanlen

Sound Field Reproduction

Humans have an amazing ability to localize sounds. Placing virtual sound sources around a listener can play an important role in critical applications such as air-traffic control and pilot early warning systems. Immersive 3D audio provides a means of achieving this, although current systems are notoriously nonrobust. The fundamental problem underlying 3D audio, and many other acoustic problems, is to control the sound field in an extended region of space. This project aims to solve the problem of recording and reproducing an arbitrary sound field in reverberant rooms.

Contact: Dr Thushara Abhayapala

Optimal Multi-Access Coding for Finite Resource Allocation

Multi-access coding has grown in popularity with the advent of code-division-multi-access (CDMA) mobile handsets. While similar concepts are found in frequency-division (FDMA), time-division (TDMA) and even spatial division on a granular scale no coherent theory of multi-user separation given finite resources has been addressed in detail. Given a fixed frequency bandwidth, a finite amount of time and a finite region of space, how should we design codes which separate several users guaranteeing orthogonality in a fading, frequency-selective, spatially diverse channel? An holistic approach for the design methodology is required, which would include current division-multi-access techniques as special cases.

Contact: Dr Leif Hanlen

Next Generation Broadband Wireless and Wireline Systems

Broadband wireless systems are faced with many problems, inherently from the wireless channel, which increase cost and performance. The aim of this research is to investigate signal processing methods to overcome these problems. Advanced techniques using multi-carrier systems like OFDM and OFDMA are investigated from a practical point of view. These can be applied to future standards for both ADSL (Broadband over telephone cables) or Mobile Wireless internet standards. Techniques for performance improvement like multiple antennas (MIMO) and advanced receiver approaches can be used to overcome communication limitations. The project will consider practical issues and develops techniques that can be realised in hardware. This project aims to develop suitable techniques to improve the performance of such a system utilizing the spatial and frequency diversity.

Contact: Dr Mark Reed

Multiuser Systems and Multiaccess Technologies

Most wireless communication systems rely on multi-access methods to communicate over a shared medium. The segmentation of this medium and the methods one can use to reduce interference and approach the capacity of these channels is a continuing area of research. Co-ordinated schemes such as time-division and frequency-division multiple access rely on closed loop timing control and this is an additional overhead to the system. Systems based on code-division multiple-access allow more uncoordinated system design but multiple access interference is present. Space-division multiple-access allows the exploitation of spatial dimension and is realized by the use of antenna array systems.

One example are receiver designs which cope with a many to one multiple access channel (mobiles to base station) and are called multiuser detectors (MUD). These receivers utilize information from all users to reduce the received interference. These techniques can improve cell size and system capacity, reducing costs to service providers and end users. Efficient designs of these receivers are still not in use and more work needs to be done to solve outstanding problems.

Contact: Dr Mark Reed